An AI data center can spend months being assembled before it starts serving customers. Nvidia wants to cut that wait dramatically.
At GTC Berlin on Oct. 21, Nvidia introduced DSX POD, a reference design it says could reduce onsite deployment at prepared sites from six months to as little as 14 days. That timeline excludes the work needed to prepare the site.
For systems integrators and infrastructure resellers, the proposed timeline makes site readiness and clear support responsibilities essential before equipment arrives.
Factory assembly reduces work at the deployment site
Built around the Vera Rubin computing platform, DSX POD uses a reference architecture covering the supporting networking, storage, power and cooling systems. Manufacturers prepare modular containers with key components already installed rather than assembling most of the infrastructure inside a conventional data hall.
Under the technical design, factories complete internal cabling and test network connections as well as cooling integrity before shipment. Site preparation can proceed in parallel while those checks take place.
A standardized 800-volt direct-current input connects the core containers to AI data center power infrastructure. Local teams must still provide the upstream electrical supply and link liquid-cooling equipment to the facility. External networking and final commissioning take place after delivery.
Dell and HPE prepare partner-built offerings
Instead of manufacturing complete facilities itself, Nvidia provides the design that hardware manufacturers and container specialists will use to build commercial systems. Initial offerings are expected to become available worldwide in early 2027, with several major suppliers outlining their roles in the rollout:
- Dell Technologies is developing a Modular Data Center with Nvidia DSX POD, extending its AI Factory portfolio with integration and testing alongside deployment and lifecycle support.
- HPE plans to support the architecture, drawing on its experience with existing AI Mod Pod systems.
- Schneider Electric has introduced RapidScale Reference Design 115, a prefabricated blueprint combining electrical and cooling infrastructure for high-density Vera Rubin facilities. It represents an initial step toward DSX POD support rather than a finished commercial offering.
Each supplier’s final system configuration will provide a clearer picture of how the offerings compare when they reach the market.
What partners should establish before quoting a project
Factory integration could change how systems integrators and infrastructure VARs scope AI projects. Moving assembly upstream may reduce some onsite installation work while increasing the relative importance of site engineering, commissioning, and lifecycle support. The effect on revenue and margins will depend on each supplier’s delivery model and the services partners retain.
Before committing to a deployment schedule, partners should confirm who is responsible for electrical capacity, cooling connections, external networking, and site acceptance. Power access delays or unfinished upgrades could prevent factory-tested equipment from entering service. Nvidia’s proposed 14-day window depends on those supporting systems being ready.
Partners should also define how the equipment will be tested after delivery and what constitutes customer acceptance. Factory validation does not replace checks for transport damage, verification of site connections, or commissioning under operating conditions.
Support agreements need similar clarity. Partners offering ongoing monitoring should establish who diagnoses faults, supplies replacement hardware, handles power or cooling failures, and coordinates escalation across suppliers. A customer-facing service commitment should reflect the response times and coverage available from those providers.
For firms evaluating AI infrastructure opportunities, the practical takeaway is to price the complete delivery and support obligation. As offerings arrive in 2027, partners should compare engineering effort, commissioning requirements, and ongoing service costs alongside the advertised deployment window.
More news: Tenstorrent’s Galaxy expansion gives systems integrators another AI computing platform to evaluate beyond established GPU offerings.